-
Notifications
You must be signed in to change notification settings - Fork 84
Add openai.chat_completions package to support OSS models
#156
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Open
dangusev
wants to merge
23
commits into
main
Choose a base branch
from
feature/baseten
base: main
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Open
Changes from 16 commits
Commits
Show all changes
23 commits
Select commit
Hold shift + click to select a range
7eaa521
LLM: always get the Conversation from the agent instead of attaching …
dangusev 6993bd5
LLM: set None instructions by default
dangusev 36daca8
Support for video LLMs hosted on Baseten
dangusev 840fd35
LLM: add method `set_conversation` to be called by the agent after it…
dangusev fb7befc
LLM: Fix type hints
dangusev 4639df2
Update uv.lock
dangusev 6f353cc
Fixed mypy
dangusev 5620334
Improve readme
Nash0x7E2 736ff77
Basic example
Nash0x7E2 39da564
LLM: add method `set_conversation` to be called by the agent after it…
dangusev f3d5b11
Rename models to ChatCompletionsLLM and ChatCompletionsVLM and move t…
dangusev 6f70b14
Fix after rebase
dangusev 3b35983
Add ChatCompletionsVLM to openai.__init__
dangusev 1bc042e
Test cleanup
dangusev ae28cce
fix ruff
dangusev 7d33615
Fix mypy
dangusev ff4508b
Fix docstring
dangusev 73b1863
Fix docstring
dangusev 83442ca
Code cleanup
dangusev 42c7d17
Fix typos
dangusev 4508013
Update README.md
dangusev e6767f4
Merge branch 'main' into feature/baseten
dangusev bd6b4cc
Fix fps forwarding
dangusev File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Some comments aren't visible on the classic Files Changed page.
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,27 +1,63 @@ | ||
| """Video frame utilities.""" | ||
|
|
||
| import io | ||
|
|
||
| import av | ||
| from PIL.Image import Resampling | ||
|
|
||
|
|
||
| def ensure_even_dimensions(frame: av.VideoFrame) -> av.VideoFrame: | ||
| """ | ||
| Ensure frame has even dimensions for H.264 yuv420p encoding. | ||
| Crops by 1 pixel if width or height is odd. | ||
| """ | ||
| needs_width_adjust = frame.width % 2 != 0 | ||
| needs_height_adjust = frame.height % 2 != 0 | ||
|
|
||
| if not needs_width_adjust and not needs_height_adjust: | ||
| return frame | ||
|
|
||
| new_width = frame.width - (1 if needs_width_adjust else 0) | ||
| new_height = frame.height - (1 if needs_height_adjust else 0) | ||
|
|
||
| cropped = frame.reformat(width=new_width, height=new_height) | ||
| cropped.pts = frame.pts | ||
| if frame.time_base is not None: | ||
| cropped.time_base = frame.time_base | ||
|
|
||
| return cropped | ||
|
|
||
|
|
||
| def frame_to_jpeg_bytes( | ||
| frame: av.VideoFrame, target_width: int, target_height: int, quality: int = 85 | ||
| ) -> bytes: | ||
| """ | ||
| Convert a video frame to JPEG bytes with resizing. | ||
| Args: | ||
| frame: an instance of `av.VideoFrame`. | ||
| target_width: target width in pixels. | ||
| target_height: target height in pixels. | ||
| quality: JPEG quality. Default is 85. | ||
| Returns: frame as JPEG bytes. | ||
| """ | ||
| # Convert frame to a PIL image | ||
| img = frame.to_image() | ||
|
|
||
| # Calculate scaling to maintain aspect ratio | ||
| src_width, src_height = img.size | ||
| # Calculate scale factor (fit within target dimensions) | ||
| scale = min(target_width / src_width, target_height / src_height) | ||
| new_width = int(src_width * scale) | ||
| new_height = int(src_height * scale) | ||
|
|
||
| # Resize with aspect ratio maintained | ||
| resized = img.resize((new_width, new_height), Resampling.LANCZOS) | ||
|
|
||
| # Save as JPEG with quality control | ||
| buf = io.BytesIO() | ||
| resized.save(buf, "JPEG", quality=quality, optimize=True) | ||
| return buf.getvalue() | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,120 @@ | ||
| # Qwen3-VL hosted on Baseten | ||
| Qwen3-VL is the latest open-source Video Language Model (VLM) from Alibaba. | ||
| This plugin allows developers to easily run the model hosted on [Baseten](https://www.baseten.co/) with Vision Agents. | ||
| The model accepts text and video and responds with text vocalised with the TTS service of your choice. | ||
|
|
||
| ## Features | ||
|
|
||
| - **Video understanding**: Automatically buffers and forwards video frames to Baseten-hosted VLM models | ||
| - **Streaming responses**: Supports streaming text responses with real-time chunk events | ||
| - **Frame buffering**: Configurable frame rate and buffer duration for optimal performance | ||
| - **Event-driven**: Emits LLM events (chunks, completion, errors) for integration with other components | ||
|
|
||
| ## Installation | ||
|
|
||
| ```bash | ||
| uv add vision-agents[openai] | ||
| ``` | ||
|
|
||
| ## Quick Start | ||
|
|
||
| ```python | ||
| from vision_agents.core import Agent, User | ||
| from vision_agents.plugins import openai, getstream, deepgram, elevenlabs, vogent | ||
|
|
||
| async def create_agent(**kwargs) -> Agent: | ||
| # Initialize the Baseten VLM | ||
| # The api key and base url can be passed via OPENAI_API_KEY and OPENAI_BASE_URL environment variables. | ||
| llm = openai.ChatCompletionsVLM(model="qwen3vl") | ||
|
|
||
| # Create an agent with video understanding capabilities | ||
| agent = Agent( | ||
| edge=getstream.Edge(), | ||
| agent_user=User(name="Video Assistant", id="agent"), | ||
| instructions="You're a helpful video AI assistant. Analyze the video frames and respond to user questions about what you see.", | ||
| llm=llm, | ||
| stt=deepgram.STT(), | ||
| tts=elevenlabs.TTS(), | ||
| turn_detection=vogent.TurnDetection(), | ||
| processors=[], | ||
| ) | ||
| return agent | ||
|
|
||
| async def join_call(agent: Agent, call_type: str, call_id: str, **kwargs) -> None: | ||
| await agent.create_user() | ||
| call = await agent.create_call(call_type, call_id) | ||
|
|
||
| with await agent.join(call): | ||
| # The agent will automatically process video frames and respond to user input | ||
| await agent.finish() | ||
| ``` | ||
|
|
||
| ## Configuration | ||
|
|
||
| ### Environment Variables | ||
|
|
||
| - **`OPENAI_API_KEY`**: Your Baseten API key (required) | ||
| - **`OPENAI_BASE_URL`**: The base URL for your Baseten API endpoint (required) | ||
dangusev marked this conversation as resolved.
Show resolved
Hide resolved
|
||
|
|
||
| ### Initialization Parameters | ||
|
|
||
| ```python | ||
| openai.ChatCompletionsVLM( | ||
| model: str, # Baseten model name (e.g., "qwen3vl") | ||
| api_key: Optional[str] = None, # API key (defaults to OPENAI_API_KEY env var) | ||
| base_url: Optional[str] = None, # Base URL (defaults to OPENAI_BASE_URL env var) | ||
| fps: int = 1, # Frames per second to process (default: 1) | ||
| frame_buffer_seconds: int = 10, # Seconds of video to buffer (default: 10) | ||
| client: Optional[AsyncOpenAI] = None, # Custom OpenAI client (optional) | ||
| ) | ||
| ``` | ||
|
|
||
| ### Parameters | ||
|
|
||
| - **`model`**: The name of the Baseten-hosted model to use. Must be a vision-capable model. | ||
| - **`api_key`**: Your Baseten API key. If not provided, reads from `OPENAI_API_KEY` environment variable. | ||
| - **`base_url`**: The base URL for Baseten API. If not provided, reads from `OPENAI_BASE_URL` environment variable. | ||
| - **`fps`**: Number of video frames per second to capture and send to the model. Lower values reduce API costs but may miss fast-moving content. Default is 1 fps. | ||
| - **`frame_buffer_seconds`**: How many seconds of video to buffer. Total buffer size = `fps * frame_buffer_seconds`. Default is 10 seconds. | ||
| - **`client`**: Optional pre-configured `AsyncOpenAI` client. If provided, `api_key` and `base_url` are ignored. | ||
|
|
||
| ## How It Works | ||
|
|
||
| 1. **Video Frame Buffering**: The plugin automatically subscribes to video tracks when the agent joins a call. It buffers frames at the specified FPS for the configured duration. | ||
|
|
||
| 2. **Frame Processing**: When responding to user input, the plugin: | ||
| - Converts buffered video frames to JPEG format | ||
| - Resizes frames to 800x600 (maintaining aspect ratio) | ||
| - Encodes frames as base64 data URLs | ||
|
|
||
| 3. **API Request**: Sends the conversation history (including system instructions) along with all buffered frames to the Baseten model. | ||
|
|
||
| 4. **Streaming Response**: Processes the streaming response and emits events for each chunk and completion. | ||
|
|
||
| ## Events | ||
|
|
||
| The plugin emits the following events: | ||
|
|
||
| - **`LLMResponseChunkEvent`**: Emitted for each text chunk in the streaming response | ||
| - **`LLMResponseCompletedEvent`**: Emitted when the response stream completes | ||
| - **`LLMErrorEvent`**: Emitted if an API request fails | ||
|
|
||
| ## Requirements | ||
|
|
||
| - Python 3.10+ | ||
| - `openai>=2.5.0` | ||
| - `vision-agents` (core framework) | ||
| - Baseten API key and base URL | ||
|
|
||
| ## Notes | ||
|
|
||
| - **Frame Rate**: The default FPS of 1 is optimized for VLM use cases. Higher FPS values will increase API costs and latency. | ||
| - **Frame Size**: Frames are automatically resized to 800x600 pixels while maintaining aspect ratio to optimize API payload size. | ||
| - **Buffer Duration**: The 10-second default buffer provides context for the model while keeping memory usage reasonable. | ||
| - **Tool Calling**: Tool/function calling support is not yet implemented (see TODOs in code). | ||
|
|
||
| ## Troubleshooting | ||
|
|
||
| - **No video processing**: Ensure the agent has joined a call with video tracks available. The plugin automatically subscribes to video when tracks are added. | ||
| - **API errors**: Verify your `OPENAI_API_KEY` and `OPENAI_BASE_URL` are set correctly and the model name is valid. | ||
| - **High latency**: Consider reducing `fps` or `frame_buffer_seconds` to decrease the number of frames sent per request. | ||
Empty file.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,21 @@ | ||
| [project] | ||
| name = "qwen3-vl-example" | ||
| version = "0.1.0" | ||
| description = "Example using Qwen3 VL hosted on Baseten with Vision Agents" | ||
| requires-python = ">=3.10" | ||
| dependencies = [ | ||
| "vision-agents", | ||
| "vision-agents-plugins-openai", | ||
| "vision-agents-plugins-getstream", | ||
| "vision-agents-plugins-deepgram", | ||
| "vision-agents-plugins-elevenlabs", | ||
| "python-dotenv", | ||
| ] | ||
|
|
||
| [tool.uv.sources] | ||
| vision-agents = { workspace = true } | ||
| vision-agents-plugins-openai = { workspace = true } | ||
| vision-agents-plugins-elevenlabs = { workspace = true } | ||
| vision-agents-plugins-getstream = { workspace = true } | ||
| vision-agents-plugins-deepgram = { workspace = true } | ||
|
|
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Uh oh!
There was an error while loading. Please reload this page.